knitr::opts_chunk$set(echo = TRUE, message = FALSE, warning = FALSE)

1 Introduction

Mental health challenges among university students are a growing concern, with academic stress often contributing to anxiety and depression. This analysis examines the impact of exam and non-exam periods on SSRI and SNRI prescription rates across Scottish health boards from 2018 to 2023.

Key questions include: - How do prescription rates differ between exam and non-exam periods? - How do demographic factors, such as the percentage of young adults (17–25), influence trends? - Which health boards show the highest changes in prescription rates?

By integrating prescription, demographic, and geographic data, this study highlights seasonal and regional variations in mental health support needs, offering insights for targeted interventions and resource allocation.

library(janitor) 
library(here) 
library(dbplyr)
library(tidyverse)

2 Data cleaning and processing

Load and clean demographic datasets

  • HB_names dataset is processed to select only the health board code and names.
  • population_data is processed to extract the total population of each health board.
  • age_data is processed to extract the total number of young adults aged 17 to 25, the typical age range of university students. The dataset is then joined with the population_data to create a column of percentage. This would give a clear view of which health board regions to consider in the analysis.
library(stringr)

# Load and clean Health board names data
HB_names <- read_csv("C:/Data_Science/B223640/data/HB_names.csv") %>%
  clean_names() %>% 
  select(hb, hb_name) %>% 
  rename(hbt = hb) %>%
  mutate(hb_name = str_remove(hb_name, "^NHS\\s"))

# Load and clean population data
population_data <- read_csv("C:/Data_Science/B223640/data/UV103_age_health_board_census.csv", skip = 10) %>%
  rename(hb_name = "Health Board Area 2019",
         hb_population = Count) %>% 
  filter(Age == "All people" & Sex == "All people") %>% 
  select(hb_name, hb_population) %>% 
  mutate(hb_name = str_remove(hb_name, "^NHS\\s")) %>% 
  arrange(desc(hb_population))

age_data <- read.csv("C:/Data_Science/B223640/data/UV103_age_health_board_census.csv", skip = 10) %>%
  filter(str_detect(Age, "17|18|19|20|21|22|23|24|25")) %>%
  rename(hb_name = "Health.Board.Area.2019") %>% 
  group_by(hb_name) %>% 
  summarise(totalpeople = sum(Count)) %>% 
  left_join(population_data, by = "hb_name") %>% 
  mutate(percentage = totalpeople * 100 / hb_population) %>% 
  arrange(desc(percentage))

Loading and processing the exam season and non-exam season datasets from 2018 to 2023

The exam season dataset includes:

  • Semester 1 finals (November and December).
  • Semester 2 finals (April and May), as SSRI and SNRI treatments typically require at least four weeks to take effect (Ankrom, 2024).

The non-exam season dataset includes data from:

  • The first two months of the university semester (Vedhara et al., 1999):
    • Semester 1 start: September and October.
    • Semester 2 start: January and February.
  • During these periods, students typically experience lighter workloads and less stress compared to exam periods.

Processing the exam season and non-exam season datasets

  • Cleaning column names for consistency.
  • Joining with demographic datasets (HB_names and age_data)
  • Filtering the data to focus on:
    • Six health boards with a high prevalence of young adults: Lothian, Grampian, Greater Glasgow and Clyde, Tayside, Fife, Forth Valley, and Lanarkshire.
    • Six commonly prescribed SSRIs and SNRIs: Escitalopram, Sertraline, Fluoxetine, Venlafaxine, Paroxetine, and Citalopram.
library(tidyverse)

# Define column_types first to ensure consistent data parsing across all files by explicitly defining how each column should be interpreted
column_types <- cols(
  HBT = col_character(),
  DMDCode = col_character(),
  BNFItemCode = col_character(),
  BNFItemDescription = col_character(),
  PrescribedType = col_character(),
  GPPractice = col_double(),
  NumberOfPaidItems = col_double(),
  PaidQuantity = col_double(),
  GrossIngredientCost = col_double(),
  PaidDateMonth = col_double()
)

# Define a reusable function to process datasets
process_exam_nonexam_data <- function(file_paths, hb_names, age_data) {
  file_paths %>%
    map_dfr(~ read_csv(., col_types = column_types)) %>%
    clean_names() %>%
    mutate(bnf_item_description = str_remove(bnf_item_description, "[_\\s].*")) %>%
    full_join(hb_names, by = c("hbt" = "hbt")) %>%
    full_join(age_data, by = "hb_name") %>%
    filter(str_detect(bnf_item_description, "ESCITALOPRAM|SERTRALINE|FLUOXETINE|VENLAFAXINE|PAROXETINE|CITALOPRAM")) %>%
    filter(hb_name %in% c("Lothian", "Grampian", "Greater Glasgow and Clyde", "Tayside", "Fife", "Forth Valley", "Lanarkshire"))
}

# List all CSV files for exam and non-exam datasets
examfiles <- list.files("C:/Data_Science/B223640/data/examszn_18-23", pattern = "csv", full.names = TRUE)
nonexamfiles <- list.files("C:/Data_Science/B223640/data/nonexamszn_18-23", pattern = "csv", full.names = TRUE)

# Process the exam and non-exam datasets using the function
examszndata <- process_exam_nonexam_data(examfiles, HB_names, age_data)
nonexamszndata <- process_exam_nonexam_data(nonexamfiles, HB_names, age_data)

3 Processing prescription data to focus on SSRIs and SNRIs and demographic data.

Steps for Analysis and the Role of pivot_wider()

  1. Compute the prescription rates for SSRIs and SNRIs during both exam and non-exam periods by accounting for the total population in each health board.

  2. Add the percentage of individuals aged 18–25 from the demographic data to assess the influence of young adult populations on prescription trends.

  3. Calculate the percentage change in prescription rates between exam and non-exam periods to identify seasonal differences in mental health needs.

  4. Use pivot_wider() to convert the data from long to wide format, where each drug becomes a separate column:

    • Organizes Drug-Specific Data: Makes it easier to compare prescriptions across health boards.
    • Facilitates Summation: Allows row-wise calculations of total prescriptions across all drugs for each health board.
    • Prepares for Per Capita Calculations: Enables easy computation of prescriptions per capita by dividing total prescriptions by population.
    • Readable Tables: Produces a concise and clean structure for creating visually appealing summaries.

  5. Use the gt package to generate a well-formatted table that summarizes key metrics, including per capita rates, demographic data, and percentage changes, for clear presentation of findings.

# Load required libraries
library(tidyverse)
library(gt)
library(dplyr)

# Define a reusable function to process wide-format data
process_wide_data <- function(data) {
  data %>%
    mutate(hb_name = str_remove(hb_name, "^NHS\\s")) %>%
    select(hb_name, bnf_item_description, paid_date_month, hb_population, paid_quantity) %>%
    group_by(hb_name, bnf_item_description) %>%
    summarise(total_prescriptions = sum(paid_quantity), .groups = "drop") %>%
    pivot_wider(
      names_from = bnf_item_description,
      values_from = total_prescriptions,
      values_fill = list(total_prescriptions = 0)
    ) %>%
    rowwise() %>%  # Ensure row-wise summation
    mutate(total_prescriptions = sum(c_across(CITALOPRAM:VENLAFAXINE), na.rm = TRUE)) %>%
    ungroup() %>%
    full_join(age_data) %>%
    drop_na()  # Remove rows with NA values
}

# Process exam and non-exam data using the function
examszndatawide <- process_wide_data(examszndata)
nonexamszndatawide <- process_wide_data(nonexamszndata)

# Add prescription per capita to examszndatawide and nonexamszndatawide
examszndatawide <- examszndatawide %>%
  mutate(prescription_per_capita = total_prescriptions / hb_population)

nonexamszndatawide <- nonexamszndatawide %>%
  mutate(prescription_per_capita = total_prescriptions / hb_population)
# Combine data for comparison, including the percentage column
combined_prescriptions <- examszndatawide %>%
  select(hb_name, prescription_per_capita_exam = prescription_per_capita, percentage) %>%
  left_join(
    nonexamszndatawide %>%
      select(hb_name, prescription_per_capita_non_exam = prescription_per_capita),
    by = "hb_name"
  )
# Add a change column to show the percentage difference between exam and non-exam seasons
combined_prescriptions <- combined_prescriptions %>%
  mutate(
    change_percent = ((prescription_per_capita_exam - prescription_per_capita_non_exam) / prescription_per_capita_non_exam) * 100
  )

# Create a prettier table with grand_summary_rows
pretty_table <- combined_prescriptions %>%
  arrange(desc(prescription_per_capita_exam)) %>%
  gt() %>%
  tab_header(
    title = "Prescription Per Capita and Age Group Percentage",
    subtitle = "Exam Season vs Non-Exam Season (2018-2023)"
  ) %>%
  fmt_number(
    columns = vars(prescription_per_capita_exam, prescription_per_capita_non_exam, percentage, change_percent),
    decimals = 2
  ) %>%
  cols_label(
    hb_name = "Health Board",
    prescription_per_capita_exam = "Exam Season (Per Capita)",
    prescription_per_capita_non_exam = "Non-Exam Season (Per Capita)",
    percentage = "17–25 Age Group (%)",
    change_percent = "Change (%)"
  ) %>%
  tab_spanner(
    label = "Prescription Per Capita",
    columns = vars(prescription_per_capita_exam, prescription_per_capita_non_exam, change_percent)
  ) %>%
  tab_style(
    style = cell_text(weight = "bold"),
    locations = cells_column_labels(everything())
  ) %>%
  tab_style(
    style = cell_fill(color = "lightblue"),
    locations = cells_body(
      columns = vars(change_percent),
      rows = change_percent > 0
    )
  ) %>%
  tab_style(
    style = cell_fill(color = "lightpink"),
    locations = cells_body(
      columns = vars(change_percent),
      rows = change_percent < 0
    )
  ) %>%
  grand_summary_rows(
    columns = vars(prescription_per_capita_exam, prescription_per_capita_non_exam, percentage),
    fns = list(
      Avg = ~mean(., na.rm = TRUE)
    ),
    formatter = fmt_number,
    decimals = 2
  ) %>%
  tab_source_note(
    source_note = "Data sourced from 2018-2023 Prescription Records"
  )

# Print the improved table
pretty_table
Prescription Per Capita and Age Group Percentage
Exam Season vs Non-Exam Season (2018-2023)
Health Board
Prescription Per Capita
 17–25 Age Group (%)
Exam Season (Per Capita) Non-Exam Season (Per Capita) Change (%)
Greater Glasgow and Clyde 83.41 80.25 3.93 25.06
Forth Valley 82.67 78.52 5.28 21.22
Fife 80.26 74.47 7.77 21.54
Lanarkshire 72.08 72.50 −0.59 19.75
Grampian 68.26 64.58 5.70 20.83
Lothian 64.53 61.34 5.20 25.61
Tayside 61.97 56.09 10.48 22.20
Avg — 73.31 69.68 — 22.32
Data sourced from 2018-2023 Prescription Records

Key Findings

The table reveals that most health boards show higher prescription rates during exam periods, with an average increase from 69.68 to 73.31 prescriptions per capita. Tayside (+10.48%) and Fife (+7.77%) exhibit the most significant increases, reflecting heightened academic stress. In contrast, Lanarkshire shows a slight decrease (-0.59%), indicating regional variations in mental health needs or prescribing practices.

While regions with higher young adult populations, such as Greater Glasgow and Clyde (25.06%), generally have elevated prescription rates, Lothian (25.61%) does not follow this trend, suggesting alternative support systems or differing prescribing norms.

These findings highlight the need for targeted mental health interventions during exam periods, particularly in regions with significant increases, and further investigation into prescribing disparities to ensure equitable support across Scotland.

4 Processing Data for Exam and Non-Exam Seasons

Define a Function to Process the Data

The following function processes the exam and non-exam datasets: - Cleans column names. - Joins with demographic data (HB_names, population_data, and age_data). - Filters for SSRIs and SNRIs commonly prescribed for exam-related stress. - Focuses on specific health boards with high percentages of young adults. - Calculates total prescriptions and per-person prescription rates.

# Define a function to process the data
process_data <- function(data, season) {
  data %>%
    group_by(hb_name, bnf_item_description, totalpeople, percentage, hb_population) %>%
    summarise(total_prescription = sum(paid_quantity, na.rm = TRUE)) %>%
    filter(!is.na(bnf_item_description)) %>%
    mutate(
      Season = season,
      per_person = total_prescription / hb_population
    )
}

# Process exam and non-exam data
examszndata1 <- process_data(examszndata, "examseason")
nonexamszndata1 <- process_data(nonexamszndata, "nonexamseason")

# Combine the datasets
combined_data <- bind_rows(examszndata1, nonexamszndata1) %>%
  filter(!is.na(bnf_item_description))
library(forcats)
library(ggplot2)
library(plotly)

# Create the ggplot without text labels
p <- ggplot(combined_data, aes(
  x = total_prescription, 
  y = fct_reorder(bnf_item_description, total_prescription, .fun = sum, .desc = TRUE),  # Reorder drugs in descending order
  fill = Season
)) +
  geom_bar(stat = "identity", position = position_dodge(width = 0.8)) +
  labs(
    title = "Comparison of Total Prescriptions by Drug",
    x = "Total Prescriptions",
    y = "Drug",
    fill = "Season"
  ) +
  theme_minimal() +
  facet_wrap(~fct_reorder(hb_name, total_prescription, .fun = sum, .desc = TRUE), ncol = 2)  # Reorder health boards in descending order

# Make the plot interactive with only `total_prescription` and `Season` in the tooltip
interactive_plot <- ggplotly(p, tooltip = c("x", "fill"))

# Display the interactive plot
interactive_plot

Results: Seasonal Variations in SSRI and SNRI Prescriptions Across Scottish Health Boards

  • Total prescriptions for SSRIs and SNRIs differ between exam and non-exam seasons, with higher rates during exam periods, suggesting increased stress among students.

  • Greater Glasgow and Clyde and Lothian have the highest total prescriptions, correlating with larger student populations.

  • Fife, Forth Valley, and Tayside show lower totals but follow similar seasonal patterns.

  • Sertraline, Fluoxetine, and Citalopram are the most commonly prescribed drugs, while Escitalopram and Paroxetine have lower rates.

  • Exam seasons show a slight increase in prescriptions, highlighting academic stress as a driver for mental health support.

4.1 Combining and Analyzing Prescription Data Across Exam and Non-Exam Periods

Merging and Cleaning the Data This code combines the exam and non-exam datasets (examszndata and nonexamszndata) into a single dataset, joinedfulldata. Key steps include:

  • Extracting year and categorizing data into four academic periods: Semester 1 Finals, Semester 2 Finals, Semester 1 Start, and Semester 2 Start.
  • Grouping by year and period for analysis.
joinedfulldata <- examszndata %>% 
  full_join(nonexamszndata)

joinedfulldata <- joinedfulldata %>% 
   mutate(bnf_item_description = str_remove(bnf_item_description, "[_\\s].*")) %>% 
  mutate(
     year = substr(paid_date_month, 1, 4),
  period = case_when(
    substr(paid_date_month, 5, 6) %in% c("04", "05") ~ "Semester 1 Finals",
    substr(paid_date_month, 5, 6) %in% c("11", "12") ~ "Semester 2 Finals",
    substr(paid_date_month, 5, 6) %in% c("01", "02") ~ "Semester 2 Start",
    substr(paid_date_month, 5, 6) %in% c("09", "10") ~ "Semester 1 Start",
    TRUE ~ NA_character_
  )) %>% 
  group_by(year, period)